Integrative data analysis (in progress)
6 avril 2021
1 Multiblocks data analysis
1.1 The RGCCA and SGCCA approaches
(Tenenhaus et al. 2014)(https://doi.org/10.1093/biostatistics/kxu001)
(Tenenhaus and Tenenhaus 2011)(https://doi.org/10.1007/s11336-011-9206-8)
1.2 The mixOmics software
(Rohart et al. 2017)(https://doi.org/10.1371/journal.pcbi.1005752)
1.3 Application to the ProMetIS dataset
selvar.i <- 20 # number of features to select
latmx2.mset <- phenomis::reading(ProMetIS::statistics_singleomics_dir.c(),
report.c = "none")
sgccda.ls <- list()
for (gene.c in ProMetIS::genes.vc()) {
# gene.c <- "LAT"
wtgene.vc <- c("WT", gene.c)
for (tissue.c in ProMetIS::tissues.vc()) {
# tissue.c <- "plasma"
mset <- ProMetIS::subsetting(latmx2.mset,
genes.vc = wtgene.vc,
tissues.vc = tissue.c,
common_samples.l = TRUE)
# mset <- ProMetIA::abbrev_mset(mset)
preclinical.eset <- mset[["preclinical"]]
gene.fc <- factor(Biobase::pData(preclinical.eset)[, "gene"],
levels = wtgene.vc)
sex.fc <- factor(Biobase::pData(preclinical.eset)[, "sex"],
levels = ProMetIS::sex.vc())
mset <- mset[, setdiff(names(mset), "preclinical")]
exprs.ls <- MultiDataSet::as.list(mset)
## Correcting the 'sex' effect
exprs.ls <- lapply(exprs.ls,
function(exprs.mn)
t(sva::ComBat(exprs.mn, sex.fc)))
## Design matrix
# We create the design matrix. The weights were set to 0.1 between the blocks of omics and to 1 between the blocks of omics and the response block.
design.mn <- matrix(0.1,
ncol = length(exprs.ls),
nrow = length(exprs.ls),
dimnames = list(names(exprs.ls),
names(exprs.ls)))
diag(design.mn) <- 0
keep_x.ls <- lapply(names(exprs.ls),
function(set.c) rep(selvar.i, 2))
names(keep_x.ls) <- names(exprs.ls)
block_splsda <- mixOmics::block.splsda(X = exprs.ls,
Y = gene.fc,
ncomp = 2,
keepX = keep_x.ls,
design = design.mn)
gene_tissue.c <- paste0(gene.c, "_", tissue.c)
sgccda.ls[[gene_tissue.c]] <- list(gene.fc = gene.fc,
sex.fc = sex.fc,
exprs.ls = exprs.ls,
block_splsda = block_splsda)
}
}## Found 2 genes with uniform expression within a single batch (all zeros); these will not be adjusted for batch.
## Found 3 genes with uniform expression within a single batch (all zeros); these will not be adjusted for batch.
## Found 4 genes with uniform expression within a single batch (all zeros); these will not be adjusted for batch.
## Found 1 genes with uniform expression within a single batch (all zeros); these will not be adjusted for batch.
## Found 2 genes with uniform expression within a single batch (all zeros); these will not be adjusted for batch.
## Found 3 genes with uniform expression within a single batch (all zeros); these will not be adjusted for batch.
## Found 4 genes with uniform expression within a single batch (all zeros); these will not be adjusted for batch.1.3.1 Scores
plot_diablo <- lapply(sgccda.ls,
function(data_model.ls)
mixOmics::plotDiablo(data_model.ls[["block_splsda"]]))
1.3.2 Loadings
plot_loadings <- lapply(names(sgccda.ls),
function(gene_tissue.c)
mixOmics::plotLoadings(sgccda.ls[[gene_tissue.c]][["block_splsda"]],
comp = 1,
legend.color = ProMetIS::palette.vc()[c("WT", substr(gene_tissue.c, 1, 3))],
contrib = 'max', method = 'median'))
1.3.3 Plot individuals
plot_indiv <- lapply(names(sgccda.ls),
function(gene_tissue.c)
mixOmics::plotIndiv(sgccda.ls[[gene_tissue.c]][["block_splsda"]],
ind.names = TRUE,
legend = TRUE,
ellipse = TRUE,
col.per.group = ProMetIS::palette.vc()[c("WT", substr(gene_tissue.c, 1, 3))]))
plot_arrow <- lapply(names(sgccda.ls),
function(gene_tissue.c)
mixOmics::plotArrow(sgccda.ls[[gene_tissue.c]][["block_splsda"]],
ind.names = FALSE,
col = ProMetIS::palette.vc()[as.character(sgccda.ls[[gene_tissue.c]][["gene.fc"]])],
title = gene_tissue.c))
1.3.4 Plot variables
plot_var <- lapply(sgccda.ls,
function(data_model.ls)
mixOmics::plotVar(data_model.ls[["block_splsda"]],
var.names = FALSE,
style = 'graphics',
legend = TRUE))
1.3.5 Circos
circos_plot <- lapply(names(sgccda.ls),
function(gene_tissue.c)
mixOmics::circosPlot(sgccda.ls[[gene_tissue.c]][["block_splsda"]],
cutoff = 0.7,
line = TRUE,
comp = 1,
color.Y = ProMetIS::palette.vc()[c("WT",
substr(gene_tissue.c, 1, 3))]))
1.3.6 Heatmap
cim_diablo <- lapply(sgccda.ls,
function(data_model.ls)
mixOmics::cimDiablo(data_model.ls[["block_splsda"]],
comp = 1,
color.Y = ProMetIS::palette.vc()[c("WT", "LAT")]))
1.3.7 Network
net_work <- lapply(sgccda.ls,
function(data_model.ls)
mixOmics::network(data_model.ls[["block_splsda"]],
color.node = c("#1F78B4", '#33A02C'),
shape.node = c("rectangle", "rectangle"),
lty.edge = "solid", lwd.edge = 1,
cutoff = 0.75))
1.3.8 Integrating the variable selection results in the dataset
for (set.c in setdiff(ProMetIS::sets.vc(), "preclinical")) {
eset <- latmx2.mset[[set.c]]
fdata.df <- Biobase::fData(eset)
tissue.c <- unlist(strsplit(set.c, "_"))[2]
for (gene.c in ProMetIS::genes.vc()) {
block_splsda <- sgccda.ls[[paste0(gene.c, "_", tissue.c)]][["block_splsda"]]
loadings.ls <- block_splsda[["loadings"]]
loadings.vn <- loadings.ls[[set.c]][, "comp1"]
loadings.vn <- loadings.vn[abs(loadings.vn) > 0]
mixomics.vn <- numeric(nrow(fdata.df))
names(mixomics.vn) <- rownames(fdata.df)
stopifnot(all(names(loadings.vn) %in% names(mixomics.vn)))
mixomics.vn[names(loadings.vn)] <- loadings.vn
fdata.df[, paste0("mixomics_", gene.c)] <- mixomics.vn
}
Biobase::fData(eset) <- fdata.df
stopifnot(methods::validObject(eset))
latmx2.mset <- MultiDataSet::add_eset(latmx2.mset,
eset,
dataset.type = set.c,
GRanges = NA,
overwrite = TRUE,
warnings = FALSE)
}
latmx2.mset <- latmx2.mset[, ProMetIS::sets.vc()]2 Re-ordering metadata
#latmx2.mset <- ProMetIS::order_mset(latmx2.mset)
latmx2.mset <- ProMetIS:::metadata_select(latmx2.mset, step.c = "4_statistics_integrative")## Supplementary metadata written in:
## ../inst/extdata/4_statistics_integrative/metadata_supp.rdata2.0.1 Saving (not run)
phenomis::writing(latmx2.mset,
file.path(gsub(ProMetIS::data_dir.c(),
"../../ProMetIS/inst/extdata",
ProMetIS::statistics_integrative_dir.c())),
overwrite.l = TRUE,
report.c = "none")References
Rohart, Florian, Benoit Gautier, Amrit Singh, and Kim-Anh Le Cao. 2017. “MixOmics: An R Package for Omics Feature Selection and Multiple Data Integration.” Edited by Dina Schneidman. PLOS Computational Biology 13 (11): e1005752. https://doi.org/10.1371/journal.pcbi.1005752.
Tenenhaus, A., C. Philippe, V. Guillemot, K.-A. Le Cao, J. Grill, and V. Frouin. 2014. “Variable Selection for Generalized Canonical Correlation Analysis.” Biostatistics 15 (3): 569–83. https://doi.org/10.1093/biostatistics/kxu001.
Tenenhaus, Arthur, and Michel Tenenhaus. 2011. “Regularized Generalized Canonical Correlation Analysis.” Psychometrika 76 (2): 257–84. https://doi.org/10.1007/s11336-011-9206-8.